LVM Deep Dive

Introduction

The Logical Volume Manager (LVM) is the most widely used volume management framework in Linux. It provides a layer of abstraction between physical storage devices and the filesystem, enabling flexible disk management: resizing volumes on the fly, creating snapshots, striping across multiple disks, and more.

This chapter covers LVM internals: how Physical Volumes (PVs), Volume Groups (VGs), and Logical Volumes (LVs) work under the hood, LVM metadata, thin provisioning, caching, and distributed locking with lvmlockd.

LVM Architecture

graph TD
    subgraph "Logical View"
        FS["Filesystem (ext4/XFS)"]
        LV1["/dev/vg0/lv_root"]
        LV2["/dev/vg0/lv_data"]
    end
    subgraph "Volume Group (vg0)"
        VG[VG Metadata + Free Space Pool]
    end
    subgraph "Physical Volumes"
        PV1["/dev/sda2 (PV)"]
        PV2["/dev/sdb1 (PV)"]
        PV3["/dev/nvme0n1p1 (PV)"]
    end

    FS --> LV1
    FS --> LV2
    LV1 --> VG
    LV2 --> VG
    VG --> PV1
    VG --> PV2
    VG --> PV3

Physical Volumes (PV)

A Physical Volume (PV) is a block device or partition that LVM uses for storage. LVM writes metadata to the beginning of the PV and divides the rest into physical extents (PEs).

Creating a PV

# Create a physical volume
pvcreate /dev/sdb1
#  Physical volume "/dev/sdb1" successfully created.

# View PV details
pvdisplay /dev/sdb1
#  --- Physical volume ---
#  PV Name               /dev/sdb1
#  VG Name               myvg
#  PV Size               100.00 GiB / not usable 4.00 MiB
#  Allocatable           yes
#  PE Size               4.00 MiB
#  Total PE              25599
#  Free PE               12799
#  Allocated PE          12800
#  PV UUID               ABCD-1234-EFGH-5678-IJKL-9012

# PV summary
pvs
#  PV         VG   Fmt  Attr PSize    PFree
#  /dev/sda2  vg0  lvm2 a--  465.25g  100.00g
#  /dev/sdb1  vg0  lvm2 a--  100.00g   50.00g

PV Header and Metadata Area

Every PV has a header at the start of the device:

graph LR
    subgraph "Physical Volume Layout"
        HEADER["PV Header<br>Sector 0"]
        MDA["Metadata Area<br>(MDA)"]
        PE0["PE 0"]
        PE1["PE 1"]
        PE2["PE 2"]
        PEN["PE n"]
    end
    HEADER --> MDA --> PE0 --> PE1 --> PE2 --> PEN

The PV header contains:

A UUID uniquely identifying the PV
Device size
Locations of metadata areas
List of all PVs in the VG (in older formats)

# Dump PV header
pvck --dump /dev/sdb1
# Found label on /dev/sdb1 at sector 1
#  PV UUID: ABCD-1234-EFGH-5678-IJKL-9012
#  PV size: 107374182400 bytes (100.00 GiB)
#  Metadata area: start=4096 size=1048576

# Backup PV metadata
vgcfgbackup myvg
#  Volume group "myvg" successfully backed up.
# Backup stored in /etc/lvm/backup/myvg

Volume Groups (VG)

A Volume Group (VG) aggregates one or more PVs into a single storage pool. The VG manages the allocation of physical extents (PEs) to logical extents (LEs) of logical volumes.

VG Operations

# Create a VG
vgcreate myvg /dev/sdb1 /dev/sdc1
#  Volume group "myvg" successfully created

# Extend a VG (add a new PV)
vgextend myvg /dev/sdd1
#  Volume group "myvg" successfully extended

# Reduce a VG (remove a PV - must be empty first)
pvmove /dev/sdd1          # Move data off the PV
vgreduce myvg /dev/sdd1   # Remove PV from VG

# Display VG details
vgdisplay myvg
#  --- Volume group ---
#  VG Name               myvg
#  System ID
#  Format                lvm2
#  Metadata Areas        2
#  Metadata Sequence No  5
#  VG Access             read/write
#  VG Status             resizable
#  MAX LV                0
#  Cur LV                2
#  Open LV               2
#  Max PV                0
#  Cur PV                2
#  Act PV                2
#  VG Size               199.99 GiB
#  PE Size               4.00 MiB
#  Total PE              51198
#  Alloc PE / Size       38400 / 150.00 GiB
#  Free  PE / Size       12798 / 49.99 GiB
#  VG UUID               1234ABCD-5678-EFGH-9012-IJKL34567890

Physical Extent (PE) Allocation

LVM divides PVs into fixed-size chunks called Physical Extents (PEs). The default PE size is 4 MiB.

# Set PE size during VG creation (must be power of 2, 8KiB to 16GiB)
vgcreate -s 8m myvg /dev/sdb1
#  Volume group "myvg" successfully created

# Maximum VG size depends on PE size:
# PE=4MiB → max ~256 TiB per VG (with 32-bit LE count)
# PE=8MiB → max ~512 TiB
# PE=16MiB → max ~1 PiB

Logical Volumes (LV)

A Logical Volume (LV) is the virtual block device that applications and filesystems see. LVs are composed of Logical Extents (LEs), which are mapped to PEs on the underlying PVs.

Basic LV Operations

# Create a logical volume
lvcreate -L 50G -n lv_data myvg
#  Logical volume "lv_data" created.

# Create with specific number of extents
lvcreate -l 100%FREE -n lv_data myvg
#  Logical volume "lv_data" created.

# Resize an LV (online, with filesystem resize)
lvextend -L +10G /dev/myvg/lv_data
resize2fs /dev/myvg/lv_data  # For ext4
# Or for XFS:
xfs_growfs /dev/myvg/lv_data

# Reduce an LV (must unmount and shrink filesystem first)
umount /dev/myvg/lv_data
e2fsck -f /dev/myvg/lv_data
resize2fs /dev/myvg/lv_data 30G
lvreduce -L 30G /dev/myvg/lv_data

# Display LV
lvdisplay /dev/myvg/lv_data
#  --- Logical volume ---
#  LV Path                /dev/myvg/lv_data
#  LV Name                lv_data
#  VG Name                myvg
#  LV UUID                1234-5678-ABCD-EFGH-IJKL-9012-MNOP
#  LV Write Access        read/write
#  LV Creation host, time server1, 2026-07-21 10:00:00 +0800
#  LV Status              available
#  # open                 1
#  LV Size                50.00 GiB
#  Current LE             12800
#  Segments               1
#  Allocation             inherit
#  Read ahead sectors     auto
#  - currently set to     256
#  Block device           253:0

LV Segment Types

LVM supports multiple segment types for different purposes:

# Linear (default) - contiguous or fragmented allocation
lvcreate -L 50G -n lv_linear myvg

# Striped - data spread across PVs for performance
lvcreate -L 50G -n lv_striped -i 2 -I 256k myvg
# -i 2 = 2 stripes
# -I 256k = stripe size

# RAID1 - mirrored
lvcreate -L 50G -n lv_mirror --type raid1 -m 1 myvg

# RAID5
lvcreate -L 100G -n lv_raid5 --type raid5 -i 3 myvg

# RAID6
lvcreate -L 100G -n lv_raid6 --type raid6 -i 4 myvg

# View segment details
lvs -o +segtype,stripes,stripe_size
#  LV        VG   Attr       LSize   Pool Origin Data%  Type    #Str Stripe
#  lv_linear myvg -wi-a----- 50.00g                              linear    1     0
#  lv_striped myvg -wi-a----- 50.00g                              striped   2   256k
#  lv_mirror myvg rwi-a-r--  50.00g                              raid1     2     0

LVM Metadata

LVM stores metadata in two locations on each PV for redundancy:

Metadata Area (MDA)

# View metadata locations
pvdisplay -m /dev/sdb1
#  --- Physical volume ---
#  PV Name               /dev/sdb1
#  VG Name               myvg
#  ...
#  --- Physical Segments ---
#  Physical extent 0 to 12799:
#    Logical volume  /dev/myvg/lv_data
#    Logical extents 0 to 12799

# Metadata is stored as ASCII text
# Location: /etc/lvm/backup/ (automatic backups)
# Location: /etc/lvm/archive/ (archived on change)
cat /etc/lvm/backup/myvg
# # Generated by LVM2 version 2.03.16(2) (2026-07-21):
# # ...
# myvg {
#     id = "1234ABCD-5678-EFGH-9012-IJKL34567890"
#     seqno = 5
#     format = "lvm2"
#     status = ["RESIZEABLE", "READ", "WRITE"]
#     flags = []
#     extent_size = 8192       # 4 Megabytes
#     max_lv = 0
#     max_pv = 0
#     metadata_copies = 0
#
#     physical_volumes {
#         pv0 {
#             id = "ABCD-1234-EFGH-5678-IJKL-9012"
#             device = "/dev/sdb1"
#             status = ["ALLOCATABLE"]
#             flags = []
#             dev_size = 209715200  # 100 Gigabytes
#             pe_start = 2048
#             pe_count = 25599
#         }
#     }
#
#     logical_volumes {
#         lv_data {
#             id = "1234-5678-ABCD-EFGH-IJKL-9012-MNOP"
#             status = ["READ", "WRITE", "VISIBLE"]
#             flags = []
#             creation_time = 1658374800
#             creation_host = "server1"
#             segment_count = 1
#
#             segment1 {
#                 start_extent = 0
#                 extent_count = 12800  # 50 Gigabytes
#                 type = "linear"
#                 stripe_count = 1
#
#                 stripes = [
#                     "pv0", 0
#                 ]
#             }
#         }
#     }
# }

Metadata Backup and Restore

# Manual backup
vgcfgbackup myvg

# Restore metadata (CAUTION: overwrites current metadata)
vgcfgrestore myvg

# Restore to a specific PV
vgcfgrestore -f /etc/lvm/backup/myvg myvg

# Check metadata consistency
vgck myvg

Thin Provisioning

Thin provisioning allows you to overallocate storage—create LVs that are larger than the available physical space. Space is allocated on demand from a thin pool.

graph TD
    subgraph "Volume Group (100 GiB)"
        TP["Thin Pool (80 GiB)<br>Data: 60 GiB used<br>Meta: 128 MiB"]
        LV_T1["Thin LV 1 (50 GiB)<br>Used: 30 GiB"]
        LV_T2["Thin LV 2 (50 GiB)<br>Used: 20 GiB"]
        LV_T3["Thin LV 3 (40 GiB)<br>Used: 10 GiB"]
        FREE["Free: 20 GiB"]
    end

    LV_T1 --> TP
    LV_T2 --> TP
    LV_T3 --> TP
    TP --> FREE

Thin Provisioning Commands

# Create a thin pool
lvcreate -L 80G --thinpool thin_pool myvg
#  Logical volume "thin_pool" created.

# Create thin LVs (can exceed pool size!)
lvcreate -V 50G --thin -n thin_lv1 myvg/thin_pool
lvcreate -V 50G --thin -n thin_lv2 myvg/thin_pool
lvcreate -V 40G --thin -n thin_lv3 myvg/thin_pool
# Total allocated: 140 GiB, but pool is only 80 GiB!

# Check usage
lvs myvg/thin_pool
#  LV        VG   Attr       LSize  Pool   Data%  Meta%
#  thin_pool myvg twi-aotz-- 80.00g               75.00  25.00

lvs myvg
#  LV        VG   Attr       LSize  Pool       Origin Data%
#  thin_lv1  myvg Vwi-a-tz-- 50.00g thin_pool         60.00
#  thin_lv2  myvg Vwi-a-tz-- 50.00g thin_pool         40.00
#  thin_lv3  myvg Vwi-a-tz-- 40.00g thin_pool         25.00

# Monitor thin pool - ALERT when data or meta > 80%!
lvs -o +data_percent,metadata_percent myvg/thin_pool

Thin Provisioning Snapshots

Snapshots of thin LVs are very efficient—they share unchanged data with the origin:

# Create a thin snapshot (nearly instant, no space used initially)
lvcreate -s --name thin_snap1 myvg/thin_lv1

# Snapshot grows as data diverges from origin
lvs myvg/thin_snap1
#  LV          VG   Attr       LSize  Pool       Origin    Data%
#  thin_snap1  myvg Vwi---tz-k 50.00g thin_pool  thin_lv1  0.00

# Activate and use the snapshot
lvchange -ay myvg/thin_snap1
mount /dev/myvg/thin_snap1 /mnt/snapshot

LVM Caching (dm-cache)

LVM caching uses a fast device (SSD/NVMe) as a cache tier for a slower device (HDD):

graph LR
    subgraph "Cached LV"
        subgraph "Cache Layer (dm-cache)"
            SSD["Fast Device<br>(SSD/NVMe)<br>Cache"]
            HDD["Slow Device<br>(HDD)<br>Origin"]
        end
        LV["Logical Volume<br>(/dev/vg0/cached_lv)"]
    end
    LV --> SSD
    LV --> HDD

Cache Commands

# Create a cached LV
# 1. Create the origin LV on HDD
lvcreate -L 500G -n lv_slow myvg /dev/sdb1

# 2. Create cache pool on SSD
lvcreate -L 50G -n cache_pool_data myvg /dev/nvme0n1p1
lvcreate -L 1G -n cache_pool_meta myvg /dev/nvme0n1p1

# 3. Convert to cache pool
lvconvert --type cache-pool --poolmetadata myvg/cache_pool_meta myvg/cache_pool_data

# 4. Attach cache to origin
lvconvert --type cache --cachepool myvg/cache_pool_data myvg/lv_slow

# The result
lvs -o +cache_policy,cache_mode,cache_settings myvg/lv_slow
#  LV       VG   Attr       LSize   Pool          Data%  cache_policy cache_mode
#  lv_slow  myvg Cwi-a-C--- 500.00g cache_pool_data 15.00 smq          writethrough

# Cache modes:
# writethrough - writes go to both cache and origin (safe)
# writeback    - writes go to cache only (faster, risk of data loss on cache failure)

lvchange --cachemode writeback myvg/lv_slow

Snapshot Volumes

Traditional (non-thin) LVM snapshots use a copy-on-write mechanism:

# Create a snapshot (must have enough free space in VG)
lvcreate -L 10G -s -n lv_data_snap /dev/myvg/lv_data
#  Logical volume "lv_data_snap" created.

# Mount the snapshot
mount -o ro /dev/myvg/lv_data_snap /mnt/snapshot

# Check snapshot usage
lvs myvg/lv_data_snap
#  LV           VG   Attr       LSize  Pool Origin Data%
#  lv_data_snap myvg swi-a-s--- 10.00g      lv_data 25.00

# Merge snapshot back (restores origin to snapshot state)
umount /mnt/snapshot
lvconvert --merge myvg/lv_data_snap
#  Merging of snapshot myvg/lv_data_snap will occur on next activation of myvg/lv_data.

lvmlockd: Distributed Locking

In clustered environments (e.g., shared SAN storage accessed by multiple nodes), lvmlockd provides distributed locking for LVM:

graph TD
    subgraph "Node 1"
        LVM1["lvm CLI"] --> LL1["lvmlockd"]
    end
    subgraph "Node 2"
        LVM2["lvm CLI"] --> LL2["lvmlockd"]
    end
    subgraph "Lock Manager"
        DLM["DLM / sanlock"]
    end
    LL1 --> DLM
    LL2 --> DLM

Setting Up lvmlockd

# Install lvmlockd and dlm
apt install lvm2-lockd dlm  # Debian/Ubuntu
yum install lvm2-lockd dlm  # RHEL/CentOS

# Enable and start lvmlockd
systemctl enable lvmlockd
systemctl start lvmlockd

# Create a shared VG (requires a lock manager)
vgcreate --shared mysharedvg /dev/sdb1

# Or convert existing VG to shared
vgchange --lock-type dlm myvg

# Activate LV on a specific node
vgchange --lock-start mysharedvg
lvchange -ay mysharedvg/lv_data

# Stop locking on a node
vgchange --lock-stop mysharedvg

Lock Types

Lock Type	Backend	Use Case
`none`	No locking	Single node
`dlm`	DLM (Distributed Lock Manager)	Clustered VGs
`sanlock`	sanlock	Simpler setups, SAN-based

LVM Configuration

# Main config file
cat /etc/lvm/lvm.conf
# Key settings:

# Global settings
global {
    umask = 077
    use_lvmetad = 1        # Use lvmetad for caching
    use_lvmpolld = 1
    # ...
}

# Device filter (which devices to scan)
devices {
    filter = ["a|sd.*|", "a|nvme.*|", "r|.*|"]
    # Accept sd* and nvme*, reject everything else
}

# Thin pool auto-extend
activation {
    thin_pool_autoextend_threshold = 80   # Extend at 80% usage
    thin_pool_autoextend_percent = 20     # Extend by 20%
}

# Snapshot auto-extend
activation {
    snapshot_autoextend_threshold = 80
    snapshot_autoextend_percent = 20
}

lvm.conf Device Filter Examples

# Only use /dev/sda and /dev/sdb
filter = ["a|/dev/sd[ab]|", "r|.*|"]

# Exclude loop devices
filter = ["a|.*|", "r|/dev/loop.*|"]

# Only NVMe devices
filter = ["a|/dev/nvme.*|", "r|.*|"]

LVM Event Monitoring

# Monitor LVM events
dmeventd -f &

# Set up monitoring for thin pool
lvchange --monitor y myvg/thin_pool

# View monitoring status
lvs -o +monitor myvg

Common LVM Recipes

Move LV Between PVs

# Move data from one PV to another (online!)
pvmove /dev/sdb1 /dev/sdd1

# Move specific LV's extents
pvmove /dev/sdb1:0-12799 /dev/sdd1

Rename LV

lvrename myvg lv_old lv_new

Remove LV

umount /dev/myvg/lv_data
lvremove myvg/lv_data

Placement Preparation Knowledge Base